This study examines the regional data in transportation carbon emissions across China and investigates the shifting trends of the carbon emission centroid over time. Using the IPCC (2006) carbon emission calculation formula, emissions data from 30 provinces for the years 2005, 2010, 2015, and 2020 were analyzed using an Exploratory Spatial Data Analysis (ESDA) model. The Economic Centroid Model and standard deviation ellipse were applied to assess the movement of the carbon emission centroid, which was consistently located in Henan Province. Further analysis using the Kaya model identifies the key factors influencing transportation carbon emissions in Henan from 2005 to 2020. The findings offer insights into regional carbon reduction strategies and the challenges in achieving China's dual carbon goals.
The performance of Large Language Models (LLMs) is highly dependent on the input prompt, which has led to several recent studies focused on prompt optimization. Optimizing prompts for generation tasks that align with user intent is particularly challenging due to the subjective nature of these tasks and the prohibitive costs associated with the extensive human annotations required for repeated evaluations. This paper addresses the problem of optimizing prompts for LLMs in generation tasks by utilizing user pairwise comparison feedback instead of numerical scores. We propose a framework that integrates user pairwise comparison feedback into the prompt optimization process by maintaining a dynamically ranked set of prompts. In each iteration, a new prompt is generated based on previous user pairwise annotations. We evaluated our approach and an existing score-based method on 26 generation tasks based on the Big-Gen benchmark, demonstrating that our method requires only $\mathrm{O}(n)$ human annotations for n iterations, whereas the score-based method requires $\mathrm{O}(n^{2})$ annotations for the same number of iterations. Furthermore, we conducted at-test analysis, which indicated no statistically significant difference in effectiveness between our approach and the score-based method. These results suggest that our approach achieves comparable effectiveness while substantially reducing the annotation workload.
The General Data Protection Regulation (GDPR) provides data subjects with several rights ranging from data access to the erasure of personal data. Data subjects who want to use their rights must pose subject access requests (SARs) to the controllers. The practice has shown that controllers often do not properly authenticate subjects during requests, leading to potential disclosures of personal data to the wrong people. We propose an approach that provides strong security for SAR processes to prevent unwanted data leaks. The approach is also user-friendly, minimizing the hassles for users when posing SARs. We are the first to propose to use the Self-Sovereign Identity (SSI) paradigm for SARs.
Social media platforms have become essential in modern society, with usage rapidly increasing. These platforms serve various roles, from entertainment and personal expression to education and e-commerce. Users frequently share opinions, generating valuable but largely unstructured data. Businesses and governments utilize sentiment analysis on this data to understand public opinions and predict behaviors, aiding in the improvement of products, services, and policies. However, the unstructured na-ture of the data makes its analysis complex. This article provide a comprehensive evaluation of several machine learning (ML) mod-els, including Support Vector Machine (SVM), Bernoulli Naive Bayes (BernoulliNB), Random Forest (RF), Gradient Boosting (GB), K-nearest Neighbors (KNN), and Logistic Regression (LR) on Twitter data for Arabic sentiment classification. Additionally, it provides valuable insights into the preprocessing steps required to prepare the data for analysis. It also discusses the selection of effective feature extraction techniques for Saudi dialect and modern Arabic sentiment analysis to gauge their efficacy in analyzing social media data. Among various feature extraction methods, TF-IDF consistently demonstrated superior accuracy, with SVM emerging as the top-performing model at 92.4%. Finally, a public dataset focusing on customer sentiment towards Saudi banks is employed to further validate our findings and their generalizability across different domains such as government policies and commercial sectors.
In a world brimming with new products continually, novel waste types are ubiquitous. This makes current image-based garbage classification systems difficult to perform well due to the long-tailed effects of distribution of garbage types, and necessitates an urgent and efficient garbage classification with abilities of detecting new and rare wastes and class-incremental learning for environmental sustainability. Therefore, we propose a framework of Online System of Garbage Image-Oriented Intelligent Classification, Submission, and Examination, facilitating the incremental garbage classification efforts. In which, to identify novel garbage effectively, we also introduced few-shot object detection method with two key algorithms: Two-Stage Object Detection Learning Algorithm and Dynamic Query-based Incremental Few-shot Learning Algorithm. Our experiment results show that Both outperform the current existing ones in dataset, MS COCO. Then, a strategy of Class-Incremental learning based Residual Network is proposed to meet the need of new waste class-incremental learning. The experimental results support our strategy. Finally, a prototype system employed the above algorithms and the strategy is described.
The deep learning-based cardinality estimation is a hot research topic in the field of database query optimization. The SQL query featurization has a direct impact on the quality of cardinality estimation models and has received widespread attention. There are significant shortcomings in the existing SQL query featurization method. (1) Existing methods typically only consider physical query plans and ignore the semantic information of logical query plans, thus leading to information loss. (2) Existing methods often use random sampling to estimate the data distribution, which can easily introduce bias, leading to cardinality estimation that does not match the actual data distribution, even facing the 0-tuple problem, which increases the challenge of model training. This paper proposes a novel SQL query featurization method, MQPHFM, to address the above problems. The method combines physical query plans and logical query plans to fully understand the semantic features of SQL, enabling the model to obtain both the execution details and semantic logic of the SQL query at the same time. Meanwhile, histograms are used as data distribution estimation tools to fully capture the possible logical relationships between columns, thereby improving the accuracy and robustness of cardinality estimation model. Experimental results on the IMDb dataset show that the proposed method can effectively improve the accuracy of cardinality estimation model.
In this paper, the Long Short Term memory net-work (LSTM) model is applied to predict the Chinese squid price index, aiming to provide accurate price trend information forecast for investors and policy makers in the squid market. First, the historical price data of squid is collected, and the data preprocessing techniques are applied, including time series formatting, sorting and data normalization. After normalizing the data using the MinMaxScaler method to ensure the consistency of data quality and model input, the data is divided into a training set and a test set, of which eighty percent is used for model training and the remaining twenty percent is used for testing. A sequential model consisting of two layers of LSTM and one fully connected layer is constructed to capture the time and dependency characteristics of price series effectively. This LSTM model is fitted and predicted on the training set and the test set respectively to evaluate the predictive performance of the model. Finally, the prediction results of the model were quantitatively evaluated by mean square error (MSE) and mean absolute error (MAE), and the results were normalized to obtain HMAE and HMSE evaluation indexes. The experimental results show that the LSTM model can effectively predict the dynamic change of the squid price index. According to the calculation of the above evaluation indicators, compared with the RNN model, the accuracy of the prediction result of the LSTM on the squid price index is improved within the range of 50%-80%, while compared with the GRU model, the accuracy is improved within the range of 5%-20%. This result further verifies the validity and accuracy of the LSTM model.
The rapid evolution of risk landscapes in supply chains necessitates innovative management tools that enhance decision-making and process efficiency. Building on our previously proposed framework, this paper introduces a software prototype named LARD-SC (LLMs for Automated Risk Detection in Supply Chains). It utilizes Large Language Models (LLMs) for automated risk identification and assessment of risks impacting on global supply chain and integrates the Cambridge Taxonomy of Business Risks (CTBR) to categorize risk events. Additionally, it employs Neo4j, a graph database, for dynamic risk visualization. This integrated approach aims to upgrade traditional Supply Chain Risk Management (SCRM) processes by providing supply chain professionals with an intuitive, interactive tool to manage risks more effectively, thereby increasing resilience and adaptability in complex supply environments. A case study testing the functionality and impact of the prototype in a real-world scenario with Apple Inc. demonstrates its efficacy, showing a significant improvement in risk detection and management efficiency.
Federated Reinforcement learning (FRL) presents a transformative approach for leveraging Human-robot collaboration (HRC) systems by addressing critical challenges in traditional learning paradigms. This paper provides a comprehensive review of the current state of FRL technology and its potential applications within HRC systems. The adaptation of FRL in HRC system is still in its infancy. This review systematically analyses the development trends, current challenges, and future prospects of various learning approaches within HRC systems. The paper highlights the critical factors of developing a conceptual frame-work for FRL within HRC systems to fully realise the potential of FRL. This paper aims to provide valuable insights and guidance for future research efforts focused on advancing FRL technology for human-robotic collaboration.
Large language models (LLMs) serve as powerful knowledge bases, capable of embedding knowledge from text into model parameters. However, the widespread issue of hallucination and opaque black-box processes restrict their broader application in professional sectors that place high demands on safety and predictability. Conversely, knowledge graphs provide inherent superiority in the authenticity and interpretability in question answering sessions. However, building a knowledge graph in specialized fields and conduct question answering over the knowledge graph are both nontrivial problems. This paper introduces an efficient knowledge graph question answering (KGQA) approach tailored to customized domain-specific knowledge graphs. Our “extract-then-bind” KGQA method leverages the in-context learning capabilities of the LLM to extract mention and relation proposals from the query, which are then matched with nodes and edges in the knowledge graph. Experimental results demonstrate the effectiveness of our approach in a question answering task using automotive user manuals. Notably, the knowledge graph was created through an automated process. By combining this innovative KGQA technique with its corresponding knowledge graph construction method, this paper proposes an effective and reliable system for addressing domain-specific knowledge question answering. This integrated solution guarantees authenticity and traceability in question answering while significantly reducing the need for manual labour.
Carbon credits are digital assets used in the fight against climate change. Understanding the complex interplay of factors that affect the carbon trading market is essential to determining the critical influences on carbon credit prices. In this study, three methods are employed to find the key factors that affect the price of carbon credits: correlation analysis, the Maximum Information Coefficient (MIC) and the proposed Carbon Credit Multi-Factor Identification (CCMFI) model. The CCMFI integrates Random Forest regression, SHAP values, and robust ranking mechanisms, providing a powerful tool to identify key price drivers. The results reveal that energy markets (crude oil and coal) and environmental factors (CO2 and GHG emissions) significantly influence carbon credit prices. Non-ferrous metal product manufacturing (NONF), which represents product prices, further highlights the critical role of economic factors. Patents on environmental technologies (PT) are also critical drivers, highlighting the importance of green technological ad-vancements in determining carbon credit prices. European Union Allowances (EUAs) also exerted a strong influence, reflecting the impact of international carbon markets. The performance evaluation of the proposed CCMFI model resulted in an MSE of 1.44, RMSE of 1.20, MAE of 0.44, and an R2 of 0.98, which confirms its effectiveness. Our results indicate that the key factors vary depending on the method applied. Therefore, we recommend using multiple explanation techniques for more reliable results.
As the traffic on highways becomes increasingly busy, littering has become a common traffic safety issue. In this study, we propose a method for highway litter detection based on object detection technology, aiming to achieve real-time and accurate detection and analysis of litter on highways. We proposed an object detection model, which is named ABiFPCN (Attention Bidirectional Feature Pyramid Cascade Network) which is based on YOLO V5 network. Firstly, we add attention mechanism into the model and incorporate a bidirectional feature pyramid network structure to YOLO V5. Secondly, we add multi-scale cascade layer to reduce over-fitting due to exponentially vanishing positive samples. Thirdly, we propose a pooling method based on weighted aggregation to improve the performance of pooling layers in convolutional neural networks when dealing with small targets in single-background scenarios such as highway littering. Finally, we validate the model on public object detection dataset and real-world highway cameras. Experimental results demonstrate that the proposed method achieves high accuracy and reliability in highway litter detection, providing effective monitoring and warning capabilities for traffic management departments, thereby enhancing the level of highway traffic safety.
Flying ad-Hoc networks (FANETs) are the embodiment of a vital expansion in wireless communication in that they have become enablers of a variety of applications ranging from disaster management to military operations. Nonetheless, the dynamic and decentralized nature of FANETs introduces significant security challenges like sibyl attacks, hidden and exposed node problems that necessitate robust and adaptable security frameworks. The review categorizes and evaluates from the general to the granular current security solutions, encompassing secure routing protocols, intrusion detection systems (IDS), lightweight cryptographic methods, and trust management frameworks before outline the limits and research gaps of the currently available literature. The paper then suggests a novel security framework based on the dynamic integration of the advanced technologies of blockchain, edge and fog computing and enabler tools. With this novel framework, this paper aims to signpost stakeholders like academic and industrial researchers and practitioners towards innovative solutions that ensure the confidentiality, integrity, availability, non-repudiation, scalability, and performance aware operation of UAV networks in increasingly intricate and antagonistic milieus.
A hashtag on X (Twitter) can attract a range of stakeholders with diverse perceptions on the hashtag. Under-standing and considering the differences between these stake-holders' perceptions on specific hashtags can be highly valuable for policy making in both public and private sectors. This paper proposes a framework to analyse the depth and width of the perception gap between stakeholders of a hashtag using sentiment analysis and zero-shot learning on large scale Twitter data. The framework was applied to the hashtag “#NDIS”, the National Disability Insurance Scheme, Australia's pilot personalised disability service which provides funding to support people with disability. Over 155,000 tweets were collected and analysed to identify and assess the perception gap between the two most significant stakeholders (NDIS participants and service providers) on six main topics during the implementation of the NDIS. The study verified the effectiveness of the proposed framework, revealing a widespread perception gap between the two NDIS stakeholders, as well as differences in their online behaviours using Twitter. The results suggest that the NDIS policymakers and service providers should pay attention to the utilisation of social media, including hearing voice from the broader public and creating channels for transparent engagement with other NDIS stakeholders.
In the modern world today, many online services require user authentication of some kind to log in to access its features, be it social media, banking, work or entertainment. For most people, the process of completing this task is straightforward, even if occasionally challenging or time consuming. However, disabled people can face additional challenges; whether through not being able to complete a verification due to lack of speed in responding, lack of necessary information provided to them or from security vulnerabilities that become exposed because of the nature of their disability. This paper aims to address these issues with a potential solution that provides a usable system for a disabled person to be able to access services easily without compromising on security. By way of a systematic literature review, this paper reinforces the need for such a system by highlighting fundamental shortfalls, and areas where advancements can occur and thereby advocating firm business incentives for doing so. Part of this review involved research for a proposed system architecture which illustrates how an authentication system can operate for disabled people while retaining both usability and security, along with a supporting business perspective to provide a sustainable lifecycle for the system.
Biopsy is considered the “gold-standard” for disease diagnosis. In clinical practice, it is necessary to scan biopsy tissue slices into electronic image, which allows for cooperative reasoning from multiple perspectives of the induction of the morphology and structure of cells and tissues, as well as the deduction of the cross disciplinary knowledge of clinical and pathological. Multimodal image-text-pair based large models have improved the ability of generalization for computational pathological analysis, but lack the diagnostic reasoning ability. To address the above limitation, a pathology-base diagnosis method combining large models and rule-based reasoning is proposed in this paper to take both performance of generalization and reasoning into account. The reasoning rules, summarized by experts to reflect medical domain knowledge, serve as the core components for diagnostic decisions. Specifically, clinical practice guidelines are transformed into rules of Datalog which is a reasoning engine based on first-order predicate logic. Based on electronic images obtained from scanning biopsy tissue samples, features extracted from pathological analysis using pre-trained large image-text models are treated as facts of Datalog for rule- fact matching during reasoning. In order to verify the usability of our proposed method, a case study of a downstream task of liver fibrosis staging is demonstrated, and comparisons are conducted among different studies.
The recent surge in drone utilization has highlighted the critical need for high-precision positioning systems. This paper presents the development of a sophisticated positioning system that ensures centimeter-level accuracy for both indoor and outdoor drone navigation by integrating a custom-designed high-precision module. The communication range of the system is significantly extended-by more than 50%-through the incorporation of a specially engineered RF front-end chip circuit. Moreover, the application of refined precision compensation algorithms enables drones to perform autonomous landings on unstable, floating ship platforms, thereby demonstrating the viability of a fully automated drone ship landing system.
Currently, deep learning methods and models have achieved significant progress in image segmentation and have been successfully applied in various practical applications. However, common methods and models are usually only suitable for situations where the segmentation target occupies a large portion of the input image and has a very obvious difference with the background. Additionally, they require high-quality datasets with balanced categories and large quantities. In the field of industrial defect detection, defects often occupy a small area of the image and are difficult to distinguish from the background. The edges of defects are often blurry, and the differences in shape and color between defect categories are relatively small. Moreover, in industrial settings, acquiring a large amount of defect data and labeling it is very expensive. Therefore, the available training data in practical applications is often small and imbalanced. These characteristics make it difficult to train common image detection models on industrial defect datasets. To address these issues, we propose a novel network architecture based on the Unet [25] segmentation model. This architecture effectively analyzes and utilizes image features extracted by commonly used backbone networks. As a result, the model can detect defects that closely resemble the background and occupy small regions. We also introduce a practical method for training segmentation models on low-quality datasets. Compared to existing segmentation models commonly used in image segmentation, our proposed architecture demonstrates better performance across various evaluation metrics. This research aims to provide a viable solution for industrial defect detection.
